#494 — A Coin Toss for the Future

Sam Harris speaks with AI alignment researcher Ryan Greenblatt about the existential risk posed by misaligned AI. They examine a recent Hugging Face incident as a real-world example of misalignment, unpack the concept of reward hacking, and draw the crucial distinction between aligning AIs and merely controlling them. The conversation also delves into the unsettling possibility of AI 'alignment faking' and the danger of advanced systems reasoning in inscrutable 'neuralese,' which could render human oversight powerless.